Billionaire Databricks and Perplexity Co-Founder Pitches AI Researchers to Not Work for Big Tech
The billionaire co-founder of Databricks and Perplexity AI, Andy Konwinski, is singularly focused on plugging the years-long drain of talent from academia to Big Tech.
He wants to encourage academics to focus on publishing more openly available research, a reaction to the move by frontier AI companies to reduce the amount of AI research they publish as they race against one another to develop the best models and AI tools.
While Google, in particular, has long been known for its research publishing, a 2026 Stanford report noted that OpenAI, Anthropic and Google no longer disclose details about the software used to train their AI models, how much computing power they used, and the size of their training datasets — crucial points that could help researchers replicate the companies’ successes.
“There are many reasons—like fundamental, societal-level, defend-democracy reasons—that open research needs to survive,” Konwinski told me at the Association for Computing Machinery’s AI conference in San Jose. He pointed to a well-known Google research paper from 2017 that would later become the basis of today’s most popular AI models and chatbots.
Now, he said, there is an “aging generation of luminaries” that aren’t publishing research externally or sticking around in academia to “train up their replacement,” meaning there could be a future scarcity of AI talent, just as artificial intelligence technologies become more complex and crucial to national security.
Konwinski in June 2025 co-founded Laude Institute, a kind of Y Combinator for academic and research types with big ideas for breakthroughs in AI research. Laude Institute, which Konwinski put $100 million into, currently funds about 70 projects through grants of up to $10 million. One such project was Terminal-Bench, a collection of AI benchmarking tools created by researchers from Stanford and Laude Institute that’s become an industry standard for evaluating AI coding tools.
Another is Stanford’s Marin project, which is focused on coming up with more efficient, predictable ways to pre-train open source models that perform on par with frontier ones at a fraction of the cost by requiring fewer AI chips or fewer “runs” or rounds of training.
Konwinski said he chats with about one dozen PhD students every month from the top 20 universities in North America, with a particular focus on Stanford, Berkeley, MIT, and Carnegie Mellon. His pitch to them: Stay in academia as a PhD student, even if that means making less than $100,000 annually for a few years, rather than accepting a $1 million to $3 million compensation package from a tech firm. If they do that, and publish openly, Laude will fund their research so they can share technology breakthroughs with the world. Eventually, you’ll get offers for tens of millions of dollars from the same companies, he tells them.
“I can get you [researchers] to fifty million. I can get you to a hundred million,” he said. “The value you have is in your ability to make the breakthrough.”
Konwinski said he’s fighting an uphill battle as AI talent wars send compensation packages through the roof: “The trade-off‘s just getting so much harder ’cause people are having to say no to salaries that are unprecedented.”
Academia also has problems getting adequate resources to produce AI models. Universities used to need up to a $10 million budget for large AI research projects but the cost has soared to around $100 million in some cases, given the expenses linked to costly AI chips, Konwinski said.
Nvidia CEO Jensen Huang, for instance, thinks universities need $1 billion budgets for AI chips—an issue that Stanford visiting professor Anjney Midha grilled him about during a recorded class lecture last month.
“We are dying out here,” said Midha, referencing the university’s crimped ability to purchase Nvidia’s expensive graphics processing units.
“You don’t have the budget for $1 billion compute,” Huang responded. “You have to find a way to change the way you do budgeting, he said.
Meanwhile, Konwinski is doing what he can to encourage open research despite the challenges. Laude Institute is beginning to accept pledges from other technologists and said it has a commitment from Berkeley professor and Google distinguished engineer Dave Patterson.
He said open research “needs a heavyweight champion in the ring.”
“We're fostering this ecosystem of open research so that I can put forward a heavyweight contender in the arena with the closed labs, the closed models, the closed techniques.”